India offers compute at one-fifth of global prices for AI, says Yotta

Representative image

India is now one of the cheapest places in the world to run artificial intelligence (AI) workloads, with compute costs slashed to nearly one-fifth of global rates, according to homegrown data centre and cloud services provider Yotta.

The company said that thanks to a combination of cutting-edge infrastructure, graphics processing unit (GPUs) that typically cost $10–$12 per hour on hyperscalers are now available in India for under $2 per hour.

“Standard GPUs that are priced at $10 or more per hour in the US are now being offered at sub-$2 in India,” Yotta chief executive officer Sunil Gupta told Moneycontrol on the sidelines of the “AI for India Summit” at the Infosys Science Foundation, organised by AI4India and CSTEP.

He added that with government subsidies under the IndiaAI Mission, this can go even lower for public-good projects.

Yotta is among the key partners of the government-backed IndiaAI compute initiative, which aims to create a sovereign, scalable cloud ecosystem. The company hosts some of the most advanced NVIDIA infrastructure globally and is one of six certified reference architecture partners worldwide.

Yotta’s model supports everything from large-scale GPU clusters to granular, token-based pricing for early-stage startups. “Not everyone needs a full GPU. We offer fractional GPUs, API-based access, and even token-level pricing. This is critical to support startups and small AI developers who can’t afford heavy upfront investments,” Gupta said while addressing a session on: AI in industry - Going beyond PoC.

“Just as India made mobile data and payments ultra-affordable, compute for AI is now following the same trajectory,” the executive further said. “If we want to unlock the next UPI moment in AI, this is what it takes, world-class infrastructure at Indian prices.”

Also, read: Inside India’s foundational LLM mission: How Yotta is bridging the compute gap

The company is also offering air-gapped, sovereign deployments for sectors such as defence and banking, which require zero exposure to the internet. “This is something hyperscalers can’t offer. A truly sovereign stack, built and hosted in India, governed by Indian law, is the missing piece for secure AI deployments at scale.”

What makes Yotta a significant player in the overall scheme of things in the IndiaAI mission is that it will provide 50 percent of the overall GPUs earmarked under the mission. Yotta has pitched thousands of Nvidia H-100 GPUs and H-200 GPUs for the IndiaAI mission.

Shekar Sivasubramanian, CEO, AI solutions firm WadhwaniAI, said that the only issue in AI is not defining the problem, and not the technology itself.

WadhwaniAI is providing 5 million students in Gujarat AI inferencing at just 5 paise, Sivasubramanian told Moneycontrol on the sidelines of the event.